Global power demand from data centres is projected to surge 165% by 2030, according to the latest forecasts from Goldman Sachs, amounting to 3% to 4% of total worldwide electricity consumption by the end of the decade.

With tech firms all around the world breaking ground on new data centres to take advantage of the AI boom, the bank estimates that within two years, AI workloads will account for 28% of all data centre capacity, a sharp rise from 13% today.

The shift towards AI is coming at a significant cost, however, with Goldman’s study finding a $270 billion (£201.6bn) increase in ‘hyperscale’ data centre investment since 2020, resulting in a projected $736 billion (£549.7bn) of combined capital expenditures in 2025 and 2026, just among the five highest-spending US hyperscale enterprises alone.

Amazon, for example, claims it has invested over $156 billion (£116.3bn) in US data centre infrastructure since 2011, while this year alone, Microsoft is on track to spend around $80 billion (£59.6bn) to build out AI-enabled data centres. 

At Meta, Mark Zuckerberg said the company intends to spend ‘hundreds of billions’ on AI development, which includes building multi-gigawatt data centres, just one of which “covers a significant part of the footprint of Manhattan.”    

“You have the biggest companies almost living in fear of being disrupted and deploying capital to play as much offence as they’re playing defence,” explained Eric Sheridan of Goldman Sachs Research, speaking to the bank’s Exchanges podcast.

Figures from Statista show that the United States is by far the leading nation when it comes to data centre build-out. 

As of March this year, the country was home to more than 5,400 data centres, vastly outnumbering second-place Germany with just 529 and the UK at 523, despite which President Trump has mandated the acceleration of even more data centre projects, with plans to provide federal support for data centres building out more than 100 megawatts of new load.      

This massive investment in new data centres is needed simply because those from the ‘pre-AI era’ cannot cope with the increasing power demands of AI technology.

Recommended reading

• Chancellor Announces £8 Billion AWS Investment in UK Data Centres
• UK Gov: Data Centres Are Now Critical National Infrastructure
• UK’s First Quantum-secure Connection Between Data Centres Announced

According to Goldman’s report, while a cutting-edge AI system in 2022, the year ChatGPT initially launched, required eight GPUs per server, by 2027, the leading systems will require closer to 576 GPUs in a ‘filing cabinet’ sized rack, requiring something like 600 kW of power, enough for 500 homes.

“Retrofitting existing facilities to support these massive jumps in power density is becoming complex and compromised,” said Frank Long, of Goldman Sachs Global Institute. “We will need new, purpose-built AI infrastructure to power the next generation.”

Of course, if data centre power demand grows to Goldman’s predicted 92 GW by 2027, this means that emissions, which are already proving a point of contention, will inevitably become worse.

Despite the bank estimating that 40% of this power demand will be met by renewables, the remaining 60% will still be driven by natural gas, increasing global carbon emissions by a projected 215-220 million tons within the next five years alone, though over the longer term, this will likely decline as nuclear power becomes a more realistic option.